Skip to content

Add a new op embedding. #306

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
fatcat-z merged 6 commits into from
Jan 13, 2023
Merged

Add a new op embedding. #306

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
479a3d1
Add a new op embedding.
fatcat-z Jan 11, 2023
d01e591
Disable a pylint error.
fatcat-z Jan 11, 2023
df2f9ea
Fix lint issue.
fatcat-z Jan 11, 2023
c30e0c8
Merge branch 'main' into new_ops_2
justinchuby Jan 11, 2023
f80920b
Merge branch 'main' into new_ops_2
justinchuby Jan 12, 2023
78f075d
Update onnxscript/function_libs/torch_aten/ops/core.py
fatcat-z Jan 12, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 6 additions & 7 deletions onnxscript/function_libs/torch_aten/ops/core.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1511,16 +1511,15 @@ def aten_einsum(
raise NotImplementedError()


@torch_op("aten::embedding")
def aten_embedding(
weight: TensorType,
indices: TensorType,
padding_idx: int = -1,
scale_grad_by_freq: bool = False,
sparse: bool = False,
) -> TensorType:
weight: TTensor,
indices: TTensor,
**kwargs,# pylint: disable=unused-argument
Copy link
Copy Markdown
Collaborator

@justinchuby justinchuby Jan 12, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
**kwargs,# pylint: disable=unused-argument
**_,

Don’t know if ** works in onnxscript, but if it does:

Copy link
Copy Markdown
Collaborator

@justinchuby justinchuby Jan 12, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Do we need to change behavior based on padding_idx?

Copy link
Copy Markdown
Contributor Author

@fatcat-z fatcat-z Jan 12, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

According to current implementation, it will only impact Training. I think ONNX Script won't know if current export is under Training mode, so the warning about possible wrong training result should be thrown by exporter, not onnx-script.

Thoughts?

Copy link
Copy Markdown
Collaborator

@justinchuby justinchuby Jan 12, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think we should also support any training behaviors when possible.

) -> TTensor:
# embedding(Tensor weight, Tensor indices, int padding_idx=-1, bool scale_grad_by_freq=False, bool sparse=False) -> Tensor

raise NotImplementedError()
return op.Gather(weight, indices)


def aten_embedding_backward(
Expand Down
1 change: 1 addition & 0 deletions onnxscript/test/function_libs/torch_aten/ops_correctness_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -290,6 +290,7 @@ def _topk_input_wrangler(
"nn.functional.adaptive_avg_pool2d": nn_ops.aten_adaptive_avg_pool2d,
"nn.functional.adaptive_avg_pool3d": nn_ops.aten_adaptive_avg_pool3d,
"nn.functional.elu": nn_ops.aten_elu,
"nn.functional.embedding": core_ops.aten_embedding,
"nn.functional.leaky_relu": nn_ops.aten_leaky_relu,
"nn.functional.linear": nn_ops.aten_linear,
"nn.functional.logsigmoid": nn_ops.aten_log_sigmoid,
Expand Down